Interpretable pH Prediction of Nylon 66 Salt Solutions via Feature Engineering‐Optimized Artificial Neural Network

Nylon 66 is a critical engineering thermoplastic whose molecular weight and performance are strictly dependent on the equimolar ratio of adipic acid and hexamethylenediamine during salt formation, which is most effectively monitored via solution pH. However, accurate pH prediction remains challenging due to the complex nonlinear ionization behavior of weak acid‐weak base systems and the limitations of traditional theoretical models. In this work, a feature engineering‐optimized artificial neural network (ANN) model was developed for high‐precision pH prediction of nylon 66 salt solutions. Six physically meaningful features were constructed, and the performance of ANN models with 2–6 input features was systematically evaluated. The optimal six‐feature model achieved a minimum test set mean squared error (MSE) of 2.45 × 10 −5 . The shapley additive explanations (SHAP) method was employed to interpret the model, revealing that temperature (average absolute SHAP value = 0.07) and concentration squared (0.063) are the most influential features. The nonlinear contribution patterns of these features quantitatively explained the unexpected pH rebound phenomenon observed at temperatures above 28°C in concentrated solutions. This work provides an accurate and interpretable tool for pH control in industrial nylon 66 salt synthesis and offers a methodological reference for machine learning modeling of small‐sample polymer systems.